TrackMasterFItter: Add mutex locks to operator()

Temporary work around to ensure TrackMasterFitter is thread safe by adding lock in operator(). Thread-safety issues noticed when running alignment where it crashes randomly (although rarely). These issues also detected by helgrind

Tr/TrackFitter/src/TrackMasterFitter.cpp

@@ -151,8 +151,10 @@ StatusCode TrackMasterFitter::failureInfo( const std::string& comment ) const {
 //=========================================================================
 StatusCode TrackMasterFitter::operator()( LHCb::span<LHCb::Track> tracks, IGeometryInfo const& geometry,
                                           const LHCb::Tr::PID& pid ) const {
-  auto       cache = createCache();
-  StatusCode sc( StatusCode::SUCCESS );
+  // add lock to ensure thread-safety
+  std::lock_guard<std::mutex> lock( m_operatorLock );
+  auto                        cache = createCache();
+  StatusCode                  sc( StatusCode::SUCCESS );
   for ( auto& track : tracks ) {
     StatusCode sc_temp = fit_r( track, cache, geometry, pid );
     if ( sc_temp.isFailure() ) { sc = StatusCode::FAILURE; }
@@ -165,7 +167,9 @@ StatusCode TrackMasterFitter::operator()( LHCb::span<LHCb::Track> tracks, IGeome
 //=========================================================================
 StatusCode TrackMasterFitter::operator()( LHCb::Track& track, IGeometryInfo const& geometry,
                                           const LHCb::Tr::PID& pid ) const {
-  auto cache = createCache();
+  // add lock to ensure thread-safety
+  std::lock_guard<std::mutex> lock( m_operatorLock );
+  auto                        cache = createCache();
   return fit_r( track, cache, geometry, pid );
 }
 

[Gerhard Raven]

Why not put the lock inside fit_r instead? And once in fit_r, can the scope of the lock be reduced even further?